Przejdź do treści
Biblioteka skilli
AI i agenci · Analiza

fleet-status

/fleet-status

Pokazuje na jednym ekranie, tylko do odczytu, stan działającego lub właśnie zakończonego /fleet-launch: czas do terminu, księgi sesji, worktree w toku, dostarczone prace i pozostały zakres.

Repozytorium
claude ↗
Licencja
MIT
Ostatnia zmiana repozytorium
26.10.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 26.10.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Pokazuje na jednym ekranie, tylko do odczytu, stan działającego lub właśnie zakończonego /fleet-launch: czas do terminu, księgi sesji, worktree w toku, dostarczone prace i pozostały zakres.

Wyzwalacze z opisu

fleet statushowwhat has the fleet shippedwhat

Argumenty

/fleet-status [--no-runway]
Oryginalny opis (EN)

One-screen read-only readout of a running or just-finished /fleet-launch — time left on the deadline, per-session ledgers with liveness, in-flight worktrees, shipped work cross-checked against git and the merge queue, and remaining certified runway. Use when the user says 'fleet status', 'how's the fleet doing', 'what has the fleet shipped', 'what's in flight', or invokes /fleet-status.

Jak działa

  1. Reading the output
  2. What this skill must NOT do
  3. Error Handling
Sekcje z nagłówków SKILL.md, w oryginale · 80 linii, 963 słów
Podgląd SKILL.md
---
name: fleet-status
description: One-screen read-only readout of a running or just-finished /fleet-launch — time left on the deadline, per-session ledgers with liveness, in-flight worktrees, shipped work cross-checked against git and the merge queue, and remaining certified runway. Use when the user says 'fleet status', 'how's the fleet doing', 'what has the fleet shipped', 'what's in flight', or invokes /fleet-status.
argument-hint: "[--no-runway]"
---

# Fleet Status

The during-view of the fleet workflow: `/auto-prep` (before) → `/fleet-launch` (dispatch) → **`/fleet-status`** (any
time) → `/fleet-retro` (after). All logic lives in [scripts/fleet-status.sh](../../scripts/fleet-status.sh); this skill
dispatches to it and narrates. Strictly read-only — no Linear writes, no git mutations — so it is safe mid-fleet,
repeatedly, or with no fleet at all.

Run from the project the fleet works on (a worktree cwd is fine):

```bash
~/.claude/scripts/fleet-status.sh              # full readout
~/.claude/scripts/fleet-status.sh --no-runway  # skip the remaining-candidates count (the one slow section)
```

## Reading the output

- **Deadline** — remaining time, `STOPPED` (a wind-down was requested), or none. The marker's fields and who writes them
  are [`/fleet-launch`](../fleet-launch/SKILL.md) § The deadline contract; ending a fleet early is
  [`/fleet-stop`](../fleet-stop/SKILL.md).
- **Headroom** (only when it matters) — the pick-time probe `/auto` parks on (`scripts/fleet-headroom.sh`), run once per
  fleet session on its own `mods/rate-limits` history. `Headroom: THROTTLE` means that account's 5h window is within the
  reserve of full — and, on a cswap-rotating machine, that no other account has room (the `cswap=` tail says which):
  sessions park between picks and re-probe at the reset, so a ledger that has gone quiet is likely parked, not dead — wait
  it out rather than reaping. `⚠️ Headroom gate is blind` means no reading exists, so every pick fails open; enable
  `rate-limits@alienfast-claude` before the next launch. Nothing prints when every reading says `PICK-OK`.
- **Scope** (epic-scoped fleets only) — the epic, its integration branch and base, and a **member burn-down**:
  `N shipped · M remaining (K in flight: …) · A added since prep: …`, plus the shipped members with their states. The
  burn-down is a live `epic-graph.sh` read against the `members` snapshot the launch recorded, so a child filed mid-run
  appears as *added since prep*; a member resolved outside the fleet's own ledgers counts as shipped only. A marker with no
  snapshot says added-since-prep is unknown. The Runway section is then the epic's own `--root` ranking.
- **Sessions** — one row per `tmp/auto-state-*.json` ledger **from the current fleet**: membership in the marker's
  `fleet_sessions`, else a ledger whose last write predates `launch_epoch` is prior-run history. Non-members are hidden
  with a count, single-run ledgers (`mode: single`) are never rowed and get their own count, and a recorded member with no
  ledger rows as `no ledger`. Liveness comes from the session registry (`claude agents --json`), joined on the ledger's
  filename key — **the ledger's recorded pid cannot answer it**: every session in a fleet embeds the fleet-root pid, so
  siblings share one value. Three row shapes:
  - **`ALIVE (<state>)`** — the registry lists it.
  - **`dead`** — the registry is available and does not list it, or lists it as ended (`dead (ended)`: a
    `done`/`failed`/`stopped` state, or the row `claude stop` leaves with `status` and `pid` gone). Only this shape may be
    treated as genuinely gone, and only this one is retro/reap material. Paired with a `status` still `active` it has two
    causes, and only one strands anything: the session died without recording an outcome (its last issue likely sits
    claimed with a preserved worktree), or it wound down cleanly and never wrote its terminal status. The transcript tells
    them apart — a `NO-CANDIDATES`/`AUTO-HALTED` tag or a `ScheduleWakeup(stop: true)` means it finished as designed;
    `fleet-metrics.py` reports that shape as `wound down but never finalized its ledger`. A finished fleet's rows read
    `dead` by design: `hooks/auto-rewake.sh` retires every drained session at its loop end (`claude stop`, then
    `claude rm`), with their ledgers `drained`.
  - **`unknown (no registry; pid …)`** — `claude` was unavailable; **not evidence of death, never reap against it.**

  A **`no ledger`** row is a session that owns a live worktree here but has written no `auto-state` file. Its shipped work
  is *not* in the Shipped cross-check below, which is built from ledgers alone — treat the row as a gap in the ledger, not
  an idle session. Only `/auto` sessions are listed; an interactive session holding a worktree appears under In flight
  only. `/fleet-retro` recovers ledger-less sessions from transcripts.
- **In flight** — live worktrees joined with Linear state and the owning session. A worktree whose issue reads terminal
  is leftover, awaiting `/reap-worktrees`.
- **Shipped, cross-checked** — the union of the fleet's ledgers, each entry verified against commits on the integration
  branch, falling back to the merge queue: merged ✓, deferred (queued — the drainer retries), or ⚠️ recorded-but-unfound
  (a session may have died between recording and merging).
- **Failed/canceled, cross-checked** — each recorded failure or cancel joined with the issue's *current* Linear state. A
  ledger entry is a claim about that run only: a later session or an interactive pickup can resolve the issue without
  writing any ledger, so a completed-type state renders as "since shipped" history. Only ⚠️ rows need action.
- **Runway** — unblocked certified candidates remaining, with the hidden-count notes (`needs decision` / `solo` /
  `deploy-gated` / `human`) passed through verbatim. Under a `backlog` launch — every epic-scoped launch is one — the
  count includes Backlog once nothing Planned/Todo is pickable and the heading says `(backlog fallback on)`.

Surface the script's markdown as-is, then narrate only where the output flags something (⚠️ rows, deferred merges,
dead-active sessions) — what it means and what resolves it; don't repeat the tables in prose.

## What this skill must NOT do

- No writes anywhere — never "fix" a stranded claim, drain the queue, or reap a worktree from here; point at
  `/fleet-retro`, `/merge-queue`, and `/reap-worktrees`.
- No session control — starting is `/fleet-launch`, stopping is `/fleet-stop`, killing is the user's `claude agents`.

## Error Handling

- Not in a git repo → the script errors; relay it.
- `linear-cli` missing or team not inferable → the Linear joins and runway are skipped with a note; the local sections
  (deadline, sessions, worktrees, git cross-check) still render.

Wersje

W bazie nie ma innych skilli o nazwie fleet-status.

Migawka bazy z 13.11.2026 · wersja paczki 04d629b869bd (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100fleet-retro
    Robi retrospektywę zakończonej floty sesji /loop /auto: mierzy sesje skryptem fleet-metrics.py, uzgadnia dostarczone zmiany z git i Linear, raportuje ustalenia i stosuje zatwierdzone poprawki.
  • 100fleet-stop
    Łagodnie wygasza działającą flotę sesji /loop /auto przez zapisanie minionego terminu, tak by każda sesja skończyła bieżące zgłoszenie i nie brała nowych; nic nie jest zabijane.
  • 100ops-fleet
    Prawdopodobnie pokazuje pulpit floty sesji CLIProxy i stan połączonych agentów.
  • 100contract-status
    Pokazuje w CLI status kontraktu lub wszystkich kontraktów w toku: fazę, pozostały czas SLA (IST i UTC), gotowość i następne działanie.
  • 100fleet-fixtures
    Uczy narzędzie czytać ekran agentów kodujących na podstawie przechwyconych fragmentów, by poprawnie rozpoznawać ich status i pytania.
  • 100fleet-protocol
    Określa zasady pracy wielu równoległych sesji Claude Code na jednym kodzie: heartbeaty, adresowanie przez router, stany deklaracji, dyscyplinę git i gotowość PR do scalenia.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.